Sam Altman says he agrees with Amodei that “committing to having independent evaluators with employee-like access is a great idea”, and OpenAI will do the same

I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks. Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.

@sama 2026-09-12 Sam Altman

Context & Ripple Effects

OpenAI had already aligned publicly with Anthropic on red lines for military use, while Altman argued in March that democratic institutions must retain control over AI. The new commitment moves that shared safety rhetoric toward an operational mechanism: access for outsiders who can inspect adherence to safeguards.

The immediate precedent is Anthropic’s same-day commitment to permanent third-party evaluator access. OpenAI’s endorsement turns what was framed as a unilateral Anthropic measure into a practice adopted by another leading lab.

First-order effects

OpenAI will need to provide independent evaluators access comparable to employees’ access, creating an external channel for checking whether its stated safety measures are followed.
Anthropic’s evaluator-access commitment gains a direct peer benchmark as OpenAI adopts the same basic model.

Second-order effects

Independent evaluation groups gain leverage to seek comparable access terms from other frontier-model developers, because two prominent labs have publicly embraced the principle.
Safety commitments become more consequential for OpenAI and Anthropic: outside evaluators can test implementation rather than relying solely on company descriptions of safeguards.

Third-order effects

If evaluator access becomes a common frontier-lab practice, competition over safety will shift from publishing principles to making internal controls auditable by independent parties.
The pattern points toward oversight becoming trust infrastructure for frontier AI, with the practical definition of evaluator access determining how meaningful those commitments are.

The trend: Frontier AI labs are moving from voluntary safety positions toward auditable governance arrangements that give independent reviewers access to operational evidence.

Discussion

@kevinroose Kevin Roose on x
We have reached the “Sam and Dario agree on something” part of the singularity.
@dylanmatt Dylan Matthews on x
Man ... what if we make it?
@daveshapi David Shapiro on x
Here's how this plays out: It doesn't China isn't slowing down. xAI isn't slowing down. “Pacing the frontier” is fear mongering theater to justify regulatory capture and eliminate competition.
@kimmonismus @kimmonismus on x
That likely seals it. The biggest and most significant frontier labs, Anthropic and OpenAI, are in agreement on slowing things down. This means the move will likely be implemented. It remains to be seen how China will respond.
@badboyfoxy @badboyfoxy on x
Today is a day we will not forget ! Sam , and Elon agreed with Dario on the slow down of AI who else will agree to this ? definitely not china
@jun_song Jun Song on x
Like I've been saying, OpenAI and Anthropic are just acting like they hate each other. They're on the same team. Anthropic played the Bad Cop with the dirty image, while OpenAI just played the Good Cop for consumers. Behind the scenes, they've been moving together the whole time.
@ctjlewis Lewis on x
Evaluators hardly sound like slowing down or pacing the frontier so I don't care and will ignore it.
@notjazii @notjazii on x
history is being written today for the first time ever, i've seen sam agree with dario, wow just wow so far elon, sam and dario have called for slow down of ai who else is joining them?
@yuchenj_uw Yuchen Jin on x
First time I've seen Sam agree with Dario since Anthropic was founded.
@yulikay Yuli Kay on x
And now BOTH Anthropic and OpenAI CEOs are saying the AI frontier needs to slow down 👀
@boazbaraktcs Boaz Barak on x
Very proud to be at OpenAI today.
@thestalwart Joe Weisenthal on x
Any random restaurant regularly has third party inspectors checking, you know, refrigerator temperature or gas lines or raw meat handling. What's the argument against having an AI “lab” equivalent?
@tracewoodgrains Jack on x
Elon and Dario are right.
@thezvi Zvi Mowshowitz on x
Okay, that was fast, and it directly addresses the issue, so I won't fully escalate. For now. But don't let it happen again, sir. https://x.com/... [embedded post]
@tszzl Roon on x
for the skeptics in government and elsewhere: “pacing the frontier” will compress the margins of the frontier labs. it is a heavy cost imposed asymmetrically on model developers with the strongest AIs in America. by its nature, it would be a terrible regulatory capture tactic
@chaos2cured Kirk Patrick Miller on x
And one more piece of proof that these labs were always working towards the same goal. Money and power for them. “Pacing” for us. •
@darrigomelanie Melanie D'Arrigo on x
This is probably so much worse than we know about if they're collectively agreeing, and already in CYA mode.
@contextconor Conor Brennan-Burke on x
the biggest thing missing from the AI safety debate is concentration of power if a handful of companies control the most powerful intelligence, that is itself a safety failure a system isn't safe if its safety depends on trusting a few actors with that much control
@arrington @arrington on x
What about China?
@andrewcurran_ Andrew Curran on x
An uneasy alliance begins to form. Sam also did not sign the Pacing letter. None of this works without an accord with China. The summit approaches, the timing for the safety side is perfect. However. There is currently no sign that President Trump, or his advisors, are on board.
@quantumtumbler @quantumtumbler on x
Dario says pace it. Elon agrees. Now Sam agrees too. Funny how fast the supposed divide disappears when the people running the race also get to shape the rules. If you want public trust, explain yourselves and get your hands off the oversight.
@so8res Nate Soares on x
I'm glad the lab heads are moving to “pace the frontier”. I wish they directly acknowledged that they're doing this on account of the double-digit chance that artificial superintelligence would kill us all.
@krassenstein Brian Krassenstein on x
The AI boom may be about to slam into the brakes. Sam Altman. Elon Musk. Dario Amodei. Three of the biggest names in AI are now warning that we may need to slow the race toward advanced AI for 1-2 years. I agree. But Wall Street has priced in an AI boom. If the AI race suddenly s…
@jason @jason on x
All the regulations being floated by frontier model companies must be viewed through the lens of them losing tokens to Open-Source models. The timing of these regulations seems to line up perfectly with Open Source closing the gap significantly (but not completely) — and @nvidia …
@tomekkorbak Tomek Korbak on x
It's kind of remarkable how quickly “independent auditors with employee access” has gone from a pipe dream to an industry standard. More next?
@w01fe Jason Wolfe on x
I am so happy we are committing to this, I think it's a really great first step towards more robust safety, transparency, democratic decisions about risks, and laying the foundations for pacing.
@timjayas Tim Jayas on x
Dario, Sam and Elon all agree that AI is too dangerous for mankind and yet develops a a new world's strongest AI model each month btw this is also the first time i've seen Sam and Dario come together so this might be a serious situation
@afinetheorem Kevin A. Bryan on x
Very good to see Sam and Elon supporting the ideas in Dario's open letter. Pacing doesn't mean slower progress! It means transparent, safe, training, and much more careful rules on what is sufficient before we release models.
@gailcweiner Gail Weiner on x
First time they have agreed since 2021
@marcus_j_w Marcus Williams on x
One of the most important developments, maybe ever
@emollick Ethan Mollick on x
This week brought some of the clearest statements we've heard from both Anthropic & OpenAI that some form of recursive self-improvement has been achieved, though it still sounds early RSI will cause rapid gain in AI ability & the first firms to RSI may get an unsurmountable lead
@joannastern Joanna Stern on x
Time to update my “A Normal Person's Guide to How AI Could Kill Us All” https://thenewthings.com/...
@ajs Alyson Shontell on x
Sam and I also discussed that Presidents Trump and Xi could win a Nobel peace prize by coming together on global AI safety guidelines at the end of Sept. I took it one step further - President Trump could be like Bill Pullman in Independence Day, getting every country to put asid…
@leolaporte Leo Laporte on x
@MiaAI_lab More like “I agree with Dario we have to stop these open weight models before they eat our lunch. Let's get the usual morons in government to protect our business model.”
@miaai_lab Mia on x
they are all aligned on this one what happened to the Nvidia letter everyone signed onto?
@aidangomez Aidan Gomez on x
Some great ideas here from the cartel: - you need to give us employee-level access to your entire operation - if we don't think you're ‘safe’ enough, sorry we're shutting you down for ‘safety’ - China won't comply, but everyone else has to! or no chips! Brilliant stuff.
@mikeisaac Rat King on x
Sam Altman, CEO of OpenAI chimes in and agrees with his rival Dario Amodei of Anthropic about an AI slowdown https://x.com/...
NewsMax.com Solange Reyner on x
AI Bosses Agree on Slowing Development

Chronicles